Tell me more ×
Cross Validated is a question and answer site for statisticians, data analysts, data miners and data visualization experts. It's 100% free, no registration required.
up vote 4 down vote favorite
1

Having used randomForest in R to produce a fairly successful classifier is there any way to emphasise sensitivity over specificity, for example, if the cost of missing a disease is much greater than diagnosing a false positive?

share|improve this question
Looks like the bst package in R will accept weights for false positive and false negatives. I'm going to leave the question open though, It might be useful to others. – rosser Nov 25 '11 at 15:42
1  
biomedcentral.com/content/pdf/1471-2105-10-S1-S22.pdf Using random forest for reliable classification and cost-sensitive learning for medical diagnosis – rosser Nov 25 '11 at 16:01

1 Answer

up vote 2 down vote accepted

I don't actually know R package, but if you were able to decrease the cost of False Positives it should work. This is also the rationale of MetaCost by Domingos, that is implemented in WEKA. However, increasing the sensitivity you are going to decrease the specificity, naturally. From the information retrieval point of view, as long as you increase the recall the precision will decrease.

Because Random Forest use Decision Trees as base classifiers and they can output probabilities, you can decrease the cut-off that enable a tree to classify a record as positive. This will make you Random Forest more sensitive but less precise.

share|improve this answer
1  
Sure. I'll add that in case of randomForest, OOB votes are in votes element of the randomForest object; for prediction, one must use predict with type="votes" or type="prob". – mbq♦ Nov 25 '11 at 22:41

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.